Superhuman AI researcher forecast jumps to 2027 as Anthropic speeds up

The forecasting group behind the AI 2027 scenario now says a superhuman AI researcher will likely show up next year. All six researchers behind the update — Eli Lifland, Brendan Halstead, Daniel Kokotajlo, Romeo Dean, Thomas Larsen and Lukas Finnveden — put their individual medians between May and November 2027, assuming no substantial deliberate slowdown. Kokotajlo assigned a 50% chance by May 2027, with Lifland's and Dean's medians landing in August. The group defines the milestone as an AI that matches the research taste of the best human researchers at an AGI company and can fully automate its coding.

The catalyst was a November 5, 2026 report from METR, a nonprofit AI evaluation organization. METR spent eleven weeks digging through Anthropic's internal research materials, model performance data and employee interviews, and most staff on the project concluded that as of October 2026, Anthropic was likely improving its models on well-specified tasks at around twice the rate it managed between March 2024 and October 2025. METR guessed roughly 80% of that acceleration came from Anthropic's own use of AI, and said it wouldn't be surprised if the true figure sat between 1.3x and 2.8x. Anthropic redacted most of the informative evidence METR relied on.

Anthropic itself hinted at the shift. In an amended S-1 filed three days before the METR report, the company said its internal estimates and expectations for future models mean it “may imminently meet” its safety policy's definition of “dramatically accelerated progress attributable to AI.” It added that it has not designated any current model as crossing that threshold.

The AI Futures Project calculated the findings imply an AI-driven speed-up of roughly 1.7x in overall research progress. Even Kokotajlo's most aggressive previous setup had assigned less than a 4% probability to that level of acceleration by November 2026. “Unfortunately, it seems like the intelligence explosion will be happening in 2027 after all,” Kokotajlo wrote on X. “If they're right, that implies an AI-driven acceleration that is 5 months ahead of the AI 2027 scenario. I am scared, sad, and angry.”

The group also puts the odds that “Top-Expert-Dominating AI” — its term for AI that can automate all white collar work — arrives before 2028 at between 35% and 65%. METR said its eleven-week review found no compute, data or algorithmic bottleneck likely to halt the acceleration, though it wasn't confident in that call and Anthropic redacted its analysis of specific bottlenecks.

This isn't the first sharp revision in the field. The AI Impacts survey of machine learning researchers put median human-level machine intelligence at 2047 in 2023, up from 2060 a year earlier, and estimated a 10% chance of unaided machines outperforming humans on every task by 2027. RAND's March 2026 assessment concluded AGI forecasting “lacks resolved forecasts for calibration,” and a 2014 study of famous AI predictions found expert judgment on timelines to be poorly reliable — Newell and Simon's 1958 chess prediction took nearly four decades, with Deep Blue beating Garry Kasparov in 1997. The AI Futures Project frames its own number as conditional: if a superhuman AI researcher really lands in 2027, it says an industry-wide pause on AI R&D “should be the top priority for the AI industry, the US government, and the entire world.”